bitsnbytes - AshokBhat/ml GitHub Wiki
About
- Lightweight Wrapper: Simplifies using CUDA custom functions in Python.
- 8-Bit Optimizers: Supports optimizers like Adam and RMSProp.
- Quantization Support: Facilitates 4-bit and 8-bit operations.
- Multi-Backend Compatibility: Works with Intel, AMD, and Apple Silicon.
- Memory Efficiency: Reduces memory usage significantly during training